NVIDIA and ServiceNow Announce Autonomous AI Agents for Enterprise

Enterprise AI is moving beyond generation and reasoning into actual action. At Knowledge 2026, NVIDIA and ServiceNow expanded their autonomous AI agent collaboration in line with this shift, combining NVIDIA accelerated computing, open models, domain-specific skills, and secure agentic execution software with ServiceNow Action Fabric and AI Control Tower.

ServiceNow introduced Project Arc, a long-running, continuously evolving desktop agent for knowledge workers, developers, IT teams, and administrators. This agent connects natively with the ServiceNow AI Platform, maintaining governance, auditability, and workflow intelligence while handling complex multi-step tasks using the local file system, terminal, and installed applications.

There are three core requirements:

• Customizable open models and domain-specific skills
• Security that reduces exposure of sensitive data and systems
• Efficient AI factories and tokenomics to support operations at scale

Project Arc runs on NVIDIA OpenShell, an open-source secure runtime that enables developing and deploying autonomous agents in sandboxed, policy-based environments. ServiceNow is contributing to OpenShell, allowing enterprises to precisely define what agents can see, what tools they use, and how execution is isolated.

Models and evaluation frameworks are also being aligned. NVIDIA Agent Toolkit, NVIDIA Nemotron open models, and NVIDIA AI-Q Blueprint become building blocks for ServiceNow AI Specialists, and NOWAI-Bench is integrated with NVIDIA NeMo Gym. Within it, on EnterpriseOps-Gym, Nemotron 3 Super currently ranks first among open-source models.

On the hardware and operations side, Blackwell is central. The companies explained that Blackwell delivers more than 50x higher token throughput per watt compared to Hopper, with cost per million tokens nearly 35x lower. ServiceNow AI Control Tower is connected with NVIDIA Enterprise AI Factory validated designs, strengthening visibility and lifecycle management of AI workloads from deployment through optimization.